Discrete op amp

Kdent

Member
2020-02-10 3:42 pm
Florida
Hello everyone, I posted this in another channel but got recommended to post it here:

I recently discovered these discrete op amps. Off the top of my head Sparkos Labs and Burson Audio makes some pretty nice ones. Would any of you have some suggestions on schematics to make one of my own? Preferably to compete with higher end IC based op amps such as Burr Browns 627/637. Thanks!
 

Mark Johnson

Member
Paid Member
2011-05-27 3:27 pm
Silicon Valley
1. Google for JE-990


2. Try this query: LINK


_
 

Attachments

  • aloha.png
    aloha.png
    49.1 KB · Views: 867
The John hardy op amp is diy-able. I built a pair on breadboards and used them as a headphone amp. The design gets a bit further out of reach with each generation due to increased complexity and rarer, smaller parts. Depending on the intended use maybe the Forsell Jfet opamp or a API 2520 (both simpler). I don’t think you’ll beat modern monolithic opamps (in terms of specs at least) very easily.
 
Plenty of ideas/schematics in this very long thread : Discrete Opamp Open Design

As to whether its worthwhile to 'go discrete' it all depends on what you're after? You can soup up an AD744 with a discrete output stage which is much less complexity of build than going full discrete. Assuming that the noise of AD744 isn't a show-stopper. If you need lower noise, try a Mitsubishi M5240P - a bit tricky to find these days but its own OPS can be bypassed.
 
Off the top of my head Sparkos Labs and Burson Audio

He-he, drop them out and forget.

Would any of you have some suggestions on schematics to make one of my own?

There are nothing magical in schematics, say differential stage for common mode cancelling, current mirror as its load, common emitter as second voltage amplification stage, more or less powerful output stage. Nothing to be undisclosed.
The magic are hidden inside frequency compensation and stability/feedback depth ratio.

Preferably to compete with higher end IC based op amps such as Burr Browns 627/637.

Fortunately this are really common jokes.
The main advantage of discrete versus monolithic are components size and more thermal dissipation resulting in more idle currents.
But, when you go deeper inside that have no doubt you'll found larger parasitic capacitances and larger parasitic leaks, more unneded thermal lag and less needed thermal coupling.
All of that will anyway make your 'straightly' designed opamp to loose with monolithic in terms of gain/widebandness which will result in less feedback depth and higher distortion of any kinds.

I’m just wondering if it is worth going through the trouble of designing a discrete BJT op amp or just get an off the shelf IC based one for a headphone amplifier/gain stage for a DAC

Pick almost any IC from the shelf(preferably uncompensated ones), wire it as inverting, add BC550 and KSC2690 as 2EF follower with ~100-200 mA idle current source as load and you're at the top end of HPA.
 
Hi Sandro, retired now don't hear much from any of the old crew not many left either. You heard that Barrie passed away?

I heard, ... terrible news. I knew he was sick from the time we designed the AD8338 (~2013), but he was looking strong then. Time will get us all as some point I guess.

By the way, what's up with the avatar? It looks like an alter-ego of Bob Widlar giving the salute.

Best, Sandro
 
Pick almost any IC from the shelf(preferably uncompensated ones), wire it as inverting, add BC550 and KSC2690 as 2EF follower with ~100-200 mA idle current source as load and you're at the top end of HPA.
Singleton outputs are a tad inefficient for my tastes (and it's probably a good idea to include some base stoppers just in case), but an EF2 would certainly provide enough gain for even a TL072 not to break a sweat, and going inverting eliminates some common distortion sources. A few mA of opamp output stage Class A bias may still help (it's not like efficiency could get much worse...). You could probably go single EF after e.g. an OPA1656 (which is, after all, rated to drive headphones on its own), and these MOS input parts also tend to do very well on common-mode distortion and as such are a good choice for noninverting applications as well.

200 mA @ +/-15 V would mean frying away a solid 6 W per channel, plus power supply losses, so we're probably talking 15 W in total. I think given the cost of heatsinking these days, most people would be gladly accepting the higher complexity of a push-pull output with its associated bias spreader for a roughly halved idle current.

The advantage of Class A buffers is that they're super easy on layouts. The more efficient you get, the more you get in terms of current pulses on the supplies. So by the time you decide to go with something BUF634-ish, you better know how to keep supply inductance down. Really, there are many different ways of tackling this particular problem.

It would be interesting to know what kind of input signal and noise levels we are dealing with and what desired output noise levels are. The required total output dynamic range for a first-class headphone amplifier may approach or exceed 130 dB (6-9 Vrms out, <= 2 µV noise, you do the math), a tall order for all but the very best DACs.
Thankfully we don't need all of this at once - a good part of it is just required to make up for variations of headphone sensitivity (ranging from <95 dB / V to >125 dB/V). Which is why it is common practice to bring up levels to line level first (0 dBFS = 2 Vrms) and then provide variable attenuation and selectable gain for the headphone out, for the best real-life dynamic range under a variety of conditions. Usually that means a volume pot and jumpers / DIP switches / toggle switches, but you can also get as fancy as the RME ADI-2 DAC and do the same with a bunch of relays for switching so it can be integrated with the main volume control ("auto ref level"), with digital attenuation filling in the gaps.
 

scott wurcer

Disabled Account
2004-01-26 3:03 pm
Belmont MA

I think he's talking about board pulls.

You can put your own output stage on the Vas and tinker around with frequency compensation. The easiest thing to do is put an external PNP (2N4248, etc.) directly across the internal PNP. This part dates from before complementary processes where there were no good PNP's. This was the original idea, we actually sold a TO can with it and the PNP to a large customer in Germany making early color scanners. Rudolf Hell (1901 – 2002): Electronic engraving, typesetting and color scanning | MULTIMEDIAMAN

Considering the cost there are modern amps that should do better, if you can get board pulls at 0.15$ it's different
 
Last edited:
I checked earlier today and they go for about 10 USD a pop on DigiKey. Am I missing something here?

What Scott said. They are indeed recovered from equipment - the ones I got are all the 'premium' K-grade ones which is interesting in itself. I've used maybe 30 so far, haven't found a single failure.

The souping up I've done is to bypass the classAB OPS entirely and use a PNP EF biassed by a current source to the +ve rail. But then I wondered if a single EF was rather too heavy a load for the Vas so I downloaded the latest version of LTspice to see if the AD744 model had a pin5. Turns out it doesn't so I just looked at simulating the input impedance of a single EF vs a darlington to see what advantage that would bring. The loop gain in the audio band probably does increase with the darlington so I'm building it now to see if stability suffers from the extra phase shift the 2EF brings over a single.